update figures

experiments_SNFROM/PCA.jl

@@ -264,8 +264,20 @@ function evolve_PCA(
     #==============#
     # get initial state
     #==============#
-    U0_norm = normalizedata(U0, md.ū, md.σu)
-    p0 = P' * U0_norm[1,:]
+
+    # only out_dim == 1 supported
+    pl, Ul = begin
+        Ud_norm = normalizedata(Ud[1, :, :], md.ū, md.σu)
+        pl = P' * Ud_norm
+
+        Ul = unnormalizedata(P * pl, md.ū, md.σu)
+        Ul = reshape(Ul, 1, size(Ul)...)
+        Ul = repeat(Ul, out_dim, 1)
+
+        pl, Ul
+    end
+
+    p0 = pl[:, 1]
 
     #==============#
     # get model
@@ -316,7 +328,7 @@ function evolve_PCA(
 
     # save files
     filename = joinpath(outdir, "evolve$(case).jld2")
-    jldsave(filename; Xdata, Tdata, Udata = Ud, Upred = Up, Ppred = ps)
+    jldsave(filename; Xdata, Tdata, Udata = Ud, Upred = Up, Ppred = ps, Plrnd = pl, Ulrnd = Ul)
 
     # print error metrics
     @show sqrt(mse(Up, Ud) / mse(Ud, 0 * Ud))

experiments_SNFROM/advect_fourier1D/advect1d_compare1.jl → experiments_SNFROM/advect_fourier1D/advect1d_compare_projection.jl

@@ -1,6 +1,7 @@
 #
-using NeuralROMs
-using CairoMakie, LaTeXStrings
+using NeuralROMs, LaTeXStrings
+import CairoMakie
+import CairoMakie.Makie
 
 joinpath(pkgdir(NeuralROMs), "experiments_SNFROM", "compare.jl") |> include
 #======================================================#
@@ -15,7 +16,7 @@ device = Lux.gpu_device()
 makedata_kws = (; Ix = :, _Ib = [1,], Ib_ = [1,], _It = :, It_ = :)
 
 function compare_advect1d_l()
-    latents = [1, 3, 4, 8, 16]
+    latents = [1, 2, 3, 4, 8, 12, 16, 20]#, 24]
 
     for latent in latents
         ll = lpad(latent, 2, "0")
@@ -57,19 +58,14 @@ function compare_advect1d_l()
         postprocess_CAE(prob, datafile, modelfile_CAE; rng)
         postprocess_SNF(prob, datafile, modelfile_SNL; rng, device)
         postprocess_SNF(prob, datafile, modelfile_SNW; rng, device)
-
-        evolve_kw = (;)
-        evolve_CAE(prob, datafile, modelfile_CAE, 1; rng, evolve_kw...,)
-        evolve_SNF(prob, datafile, modelfile_SNL, 1; rng, evolve_kw..., device)
-        evolve_SNF(prob, datafile, modelfile_SNW, 1; rng, evolve_kw..., device)
     end
 
     nothing
 end
 
 #======================================================#
 function plot_compare_advect1d_l()
-    latents = [1, 2, 3, 4, 8, 16]
+    latents = [1, 2, 3, 4, 8, 12, 16, 20]#, 24]
 
     case = 1
 
@@ -79,6 +75,9 @@ function plot_compare_advect1d_l()
     e_SNW = []
 
     for latent in latents
+
+        @show latent
+
         ll = lpad(latent, 2, "0")
 
         modeldir_PCA = joinpath(@__DIR__, "model_PCA$(ll)") # traditional
@@ -96,20 +95,21 @@ function plot_compare_advect1d_l()
         file_SNL = jldopen(file_SNL)
         file_SNW = jldopen(file_SNW)
 
+        # Ulrnd
         Ud = file_PCA["Udata"]
 
-        U_PCA = file_PCA["Upred"]
-        U_CAE = file_CAE["Upred"]
-        U_SNL = file_SNL["Upred"]
-        U_SNW = file_SNW["Upred"]
+        U_PCA = file_PCA["Ulrnd"]
+        U_CAE = file_CAE["Ulrnd"]
+        U_SNL = file_SNL["Ulrnd"]
+        U_SNW = file_SNW["Ulrnd"]
 
         N = length(Ud)
         Nr = sum(abs2, Ud) / N |> sqrt
 
-        er_PCA = sum(abs2, U_PCA - Ud) / N / Nr
-        er_CAE = sum(abs2, U_CAE - Ud) / N / Nr
-        er_SNL = sum(abs2, U_SNL - Ud) / N / Nr
-        er_SNW = sum(abs2, U_SNW - Ud) / N / Nr
+        er_PCA = sum(abs2, U_PCA - Ud) / N / Nr |> sqrt
+        er_CAE = sum(abs2, U_CAE - Ud) / N / Nr |> sqrt
+        er_SNL = sum(abs2, U_SNL - Ud) / N / Nr |> sqrt
+        er_SNW = sum(abs2, U_SNW - Ud) / N / Nr |> sqrt
 
         push!(e_PCA, er_PCA)
         push!(e_CAE, er_CAE)
@@ -118,12 +118,12 @@ function plot_compare_advect1d_l()
     end
 
     xlabel = L"N_{ROM}"
-    ylabel = L"e"
+    ylabel = L"$e_\text{proj}$"
     xlabelsize = ylabelsize = 16
 
     colors = (:orange, :green, :blue, :red, :brown,)
     styles = (:solid, :dash, :dashdot, :dashdotdot, :dot)
-    labels = (L"POD-ROM$$", L"CAE-ROM$$", L"SNFL-ROM$$", L"SNFW-ROM$$",)
+    labels = (L"POD‐ROM$$", L"CAE‐ROM$$", L"SNFL-ROM$$", L"SNFW-ROM$$",)
 
     kwl = Tuple(
         (; color = colors[i], linestyle = styles[i], label = labels[i], linewidth = 2)
@@ -135,23 +135,26 @@ function plot_compare_advect1d_l()
         for i in 1:4
     )
 
-    fig = Figure(; size = (600, 400), backgroundcolor = :white, grid = :off)
+    fig = Makie.Figure(; size = (600, 400), backgroundcolor = :white, grid = :off)
     ax  = Makie.Axis(fig[1,1]; xlabel, ylabel, xlabelsize, ylabelsize, yscale = log10)
 
-    lines!(ax, latents, e_PCA; kwl[1]...)
-    lines!(ax, latents, e_CAE; kwl[2]...)
-    lines!(ax, latents, e_SNL; kwl[3]...)
-    lines!(ax, latents, e_SNW; kwl[4]...)
+    Makie.lines!(ax, latents, e_PCA; kwl[1]...)
+    Makie.lines!(ax, latents, e_CAE; kwl[2]...)
+    Makie.lines!(ax, latents, e_SNL; kwl[3]...)
+    Makie.lines!(ax, latents, e_SNW; kwl[4]...)
 
     Makie.scatter!(ax, latents, e_PCA; kws[1]...)
     Makie.scatter!(ax, latents, e_CAE; kws[2]...)
     Makie.scatter!(ax, latents, e_SNL; kws[3]...)
     Makie.scatter!(ax, latents, e_SNW; kws[4]...)
 
-    Legend(fig[0,1], ax; orientation = :horizontal, framevisible = false, unique = true)
+    Makie.Legend(fig[0,1], ax; orientation = :horizontal, framevisible = false, unique = true)
+
+    path1 = joinpath(pkgdir(NeuralROMs), "figs", "method", "exp_nrom.pdf")
+    path2 = joinpath(@__DIR__, "compare_ll_case1.pdf")
 
-    path = joinpath(pkgdir(NeuralROMs), "figs", "method", "exp_nrom.pdf")
-    save(path, fig)
+    save(path1, fig)
+    save(path2, fig)
 
     fig
 end

experiments_SNFROM/advect_fourier1D/model_CAE01/statistics_01.txt

@@ -1,21 +1,21 @@
-Epoch [0 / 0]	TRAIN LOSS: 3.8386985e-5 || TEST LOSS: 3.8386985e-5
+Epoch [0 / 0]	TRAIN LOSS: 0.0032884406 || TEST LOSS: 0.0032884406
 #======================#
 TRAIN STATS
-R² score:             0.9999616
-MSE (mean SQR error): 3.838699e-5
-RMSE (Root MSE):      0.0061957235
-MAE (mean ABS error): 0.0042425254
-maxAE (max ABS error) 0.073194206
-Lipschitz bound:      3.2653997
+R² score:             0.9966805
+MSE (mean SQR error): 0.0032884406
+RMSE (Root MSE):      0.05734493
+MAE (mean ABS error): 0.025669334
+maxAE (max ABS error) 2.4034014
+Lipschitz bound:      5.103656
 
 #======================#
 #======================#
 TEST  STATS
-R² score:             0.9999616
-MSE (mean SQR error): 3.838699e-5
-RMSE (Root MSE):      0.0061957235
-MAE (mean ABS error): 0.0042425254
-maxAE (max ABS error) 0.073194206
-Lipschitz bound:      3.2653997
+R² score:             0.9966805
+MSE (mean SQR error): 0.003288441
+RMSE (Root MSE):      0.057344932
+MAE (mean ABS error): 0.025669334
+maxAE (max ABS error) 2.4034014
+Lipschitz bound:      5.103656
 
 #======================#

experiments_SNFROM/advect_fourier1D/model_CAE01/statistics_02.txt

@@ -1,21 +1,21 @@
-Epoch [0 / 0]	TRAIN LOSS: 1.9393403e-5 || TEST LOSS: 1.9393403e-5
+Epoch [0 / 0]	TRAIN LOSS: 0.00094652316 || TEST LOSS: 0.00094652316
 #======================#
 TRAIN STATS
-R² score:             0.9999806
-MSE (mean SQR error): 1.9393403e-5
-RMSE (Root MSE):      0.004403794
-MAE (mean ABS error): 0.0030551655
-maxAE (max ABS error) 0.06608254
-Lipschitz bound:      3.26566
+R² score:             0.9990556
+MSE (mean SQR error): 0.00094652316
+RMSE (Root MSE):      0.030765617
+MAE (mean ABS error): 0.014862344
+maxAE (max ABS error) 0.9223623
+Lipschitz bound:      5.3886147
 
 #======================#
 #======================#
 TEST  STATS
-R² score:             0.9999806
-MSE (mean SQR error): 1.9393405e-5
-RMSE (Root MSE):      0.004403794
-MAE (mean ABS error): 0.0030551655
-maxAE (max ABS error) 0.06608254
-Lipschitz bound:      3.26566
+R² score:             0.9990556
+MSE (mean SQR error): 0.00094652316
+RMSE (Root MSE):      0.030765617
+MAE (mean ABS error): 0.014862344
+maxAE (max ABS error) 0.9223623
+Lipschitz bound:      5.3886147
 
 #======================#

experiments_SNFROM/advect_fourier1D/model_CAE01/statistics_03.txt

@@ -1,21 +1,21 @@
-Epoch [0 / 0]	TRAIN LOSS: 9.435316e-6 || TEST LOSS: 9.435316e-6
+Epoch [0 / 0]	TRAIN LOSS: 0.00069662556 || TEST LOSS: 0.00069662556
 #======================#
 TRAIN STATS
-R² score:             0.9999906
-MSE (mean SQR error): 9.435316e-6
-RMSE (Root MSE):      0.003071696
-MAE (mean ABS error): 0.0021487463
-maxAE (max ABS error) 0.038793206
-Lipschitz bound:      3.2649229
+R² score:             0.9993015
+MSE (mean SQR error): 0.00069662556
+RMSE (Root MSE):      0.026393663
+MAE (mean ABS error): 0.012004193
+maxAE (max ABS error) 1.1304066
+Lipschitz bound:      5.395492
 
 #======================#
 #======================#
 TEST  STATS
-R² score:             0.9999906
-MSE (mean SQR error): 9.435317e-6
-RMSE (Root MSE):      0.003071696
-MAE (mean ABS error): 0.0021487463
-maxAE (max ABS error) 0.038793206
-Lipschitz bound:      3.2649229
+R² score:             0.9993015
+MSE (mean SQR error): 0.00069662556
+RMSE (Root MSE):      0.026393663
+MAE (mean ABS error): 0.012004193
+maxAE (max ABS error) 1.1304066
+Lipschitz bound:      5.395492
 
 #======================#

experiments_SNFROM/advect_fourier1D/model_CAE01/statistics_04.txt

@@ -1,21 +1,21 @@
-Epoch [0 / 0]	TRAIN LOSS: 7.624048e-6 || TEST LOSS: 7.624048e-6
+Epoch [0 / 0]	TRAIN LOSS: 0.000101296915 || TEST LOSS: 0.000101296915
 #======================#
 TRAIN STATS
-R² score:             0.9999924
-MSE (mean SQR error): 7.624048e-6
-RMSE (Root MSE):      0.002761168
-MAE (mean ABS error): 0.0019182552
-maxAE (max ABS error) 0.03995377
-Lipschitz bound:      3.265011
+R² score:             0.9998984
+MSE (mean SQR error): 0.00010129692
+RMSE (Root MSE):      0.010064637
+MAE (mean ABS error): 0.005334164
+maxAE (max ABS error) 0.28556395
+Lipschitz bound:      5.6360917
 
 #======================#
 #======================#
 TEST  STATS
-R² score:             0.9999924
-MSE (mean SQR error): 7.624048e-6
-RMSE (Root MSE):      0.002761168
-MAE (mean ABS error): 0.001918255
-maxAE (max ABS error) 0.03995377
-Lipschitz bound:      3.265011
+R² score:             0.9998984
+MSE (mean SQR error): 0.000101296915
+RMSE (Root MSE):      0.010064637
+MAE (mean ABS error): 0.0053341636
+maxAE (max ABS error) 0.28556395
+Lipschitz bound:      5.6360917
 
 #======================#

experiments_SNFROM/advect_fourier1D/model_CAE01/statistics_05.txt

@@ -1,21 +1,21 @@
-Epoch [0 / 0]	TRAIN LOSS: 4.4432627e-6 || TEST LOSS: 4.4432627e-6
+Epoch [0 / 0]	TRAIN LOSS: 4.514505e-5 || TEST LOSS: 4.514505e-5
 #======================#
 TRAIN STATS
-R² score:             0.9999955
-MSE (mean SQR error): 4.4432627e-6
-RMSE (Root MSE):      0.0021079048
-MAE (mean ABS error): 0.0014453864
-maxAE (max ABS error) 0.024964094
-Lipschitz bound:      3.2674487
+R² score:             0.9999548
+MSE (mean SQR error): 4.514505e-5
+RMSE (Root MSE):      0.006719006
+MAE (mean ABS error): 0.003519874
+maxAE (max ABS error) 0.1967845
+Lipschitz bound:      5.706291
 
 #======================#
 #======================#
 TEST  STATS
-R² score:             0.9999955
-MSE (mean SQR error): 4.443263e-6
-RMSE (Root MSE):      0.002107905
-MAE (mean ABS error): 0.0014453866
-maxAE (max ABS error) 0.024964094
-Lipschitz bound:      3.2674487
+R² score:             0.9999548
+MSE (mean SQR error): 4.514505e-5
+RMSE (Root MSE):      0.006719007
+MAE (mean ABS error): 0.0035198743
+maxAE (max ABS error) 0.1967845
+Lipschitz bound:      5.706291
 
 #======================#

experiments_SNFROM/advect_fourier1D/model_CAE01/statistics_06.txt

@@ -1,21 +1,21 @@
-Epoch [0 / 0]	TRAIN LOSS: 2.6286468e-6 || TEST LOSS: 2.6286468e-6
+Epoch [0 / 0]	TRAIN LOSS: 2.8584825e-5 || TEST LOSS: 2.8584825e-5
 #======================#
 TRAIN STATS
-R² score:             0.9999974
-MSE (mean SQR error): 2.6286466e-6
-RMSE (Root MSE):      0.0016213101
-MAE (mean ABS error): 0.0011346359
-maxAE (max ABS error) 0.01995939
-Lipschitz bound:      3.2689116
+R² score:             0.9999714
+MSE (mean SQR error): 2.8584822e-5
+RMSE (Root MSE):      0.0053464775
+MAE (mean ABS error): 0.0027882028
+maxAE (max ABS error) 0.16993465
+Lipschitz bound:      5.7346654
 
 #======================#
 #======================#
 TEST  STATS
-R² score:             0.9999974
-MSE (mean SQR error): 2.6286468e-6
-RMSE (Root MSE):      0.0016213102
-MAE (mean ABS error): 0.0011346359
-maxAE (max ABS error) 0.01995939
-Lipschitz bound:      3.2689116
+R² score:             0.9999714
+MSE (mean SQR error): 2.8584824e-5
+RMSE (Root MSE):      0.0053464775
+MAE (mean ABS error): 0.0027882028
+maxAE (max ABS error) 0.16993465
+Lipschitz bound:      5.7346654
 
 #======================#